Architecture of the Sema3a/Plexina4/Neuropilin Tripartite Complex ✉ ✉ Defen Lu1,5, Guijun Shang1,5, Xiaojing He2,5, Xiao-Chen Bai 3,4 & Xuewu Zhang 1,3

ARTICLE

https://doi.org/10.1038/s41467-021-23541-x OPEN Architecture of the Sema3A/PlexinA4/Neuropilin tripartite complex ✉ ✉ Defen Lu1,5, Guijun Shang1,5, Xiaojing He2,5, Xiao-chen Bai 3,4 & Xuewu Zhang 1,3

Secreted class 3 semaphorins (Sema3s) form tripartite complexes with the plexin receptor and neuropilin coreceptor, which are both transmembrane proteins that together mediate semaphorin signal for neuronal axon guidance and other processes. Despite extensive

1234567890():,; investigations, the overall architecture of and the molecular interactions in the Sema3/ plexin/neuropilin complex are incompletely understood. Here we present the cryo-EM structure of a near intact extracellular region complex of Sema3A, PlexinA4 and Neuropilin 1 (Nrp1) at 3.7 Å resolution. The structure shows a large symmetric 2:2:2 assembly in which each subunit makes multiple interactions with others. The two PlexinA4 molecules in the complex do not interact directly, but their membrane proximal regions are close to each other and poised to promote the formation of the intracellular active dimer for signaling. The structure reveals a previously unknown interface between the a2b1b2 module in Nrp1 and the Sema domain of Sema3A. This interaction places the a2b1b2 module at the top of the complex, far away from the plasma membrane where the transmembrane regions of Nrp1 and PlexinA4 embed. As a result, the region following the a2b1b2 module in Nrp1 must span a large distance to allow the connection to the transmembrane region, suggesting an essential role for the long non-conserved linkers and the MAM domain in neuropilin in the semaphorin/plexin/neuropilin complex.

1 Department of Pharmacology, University of Texas Southwestern Medical Center, Dallas, TX, USA. 2 Key Laboratory of Molecular Biophysics of the Ministry of Education, College of Life Science and Technology, Huazhong University of Science and Technology, Wuhan, China. 3 Department of Biophysics, University of Texas Southwestern Medical Center, Dallas, TX, USA. 4 Department of Cell Biology, University of Texas Southwestern Medical Center, Dallas, TX, USA. ✉ 5These authors contributed equally: Defen Lu, Guijun Shang, Xiaojing He. email: [email protected]; [email protected]

NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x

emaphorins are the largest family of guidance molecules plexins for signaling31,32. Overexpression of one plexin in some that transduce signal to control cell morphological changes, cases could overcome this requirement31,32. These observations S fi migration and directional growth. Speci cally, semaphorin suggest that different plexin family members can heterodimerize signaling plays essential roles in the development and functions of and gain extra structural and functional features, the precise the nervous system and cardiovascular system by serving as mechanisms of which are not well understood at present. guidance cues for developing neurons and blood vessels1,2. Loss- Extensive structural analyses of plexin have led to a general of-function of semaphorins leads to severe developmental defects model on how dimeric semaphorin binds and induces the in these systems and in some cases embryonic lethality1,2. dimerization and activation of plexin6,7,9,10,13,33,34. While all the Semaphorin-mediated signal also play important roles in reg- plexins use similar mechanisms of activation, there are some ulating other processes, such as wound healing, immunity, and important differences for some plexin family members. For bone development1–3. example, PlexinC1 has a distinct extracellular architecture, and Semaphorins are divided into eight classes based on sequence therefore the overall shape of the semaphorin-induced dimer of conservation. They are either cell surface attached or secreted PlexinC1 appears quite different from that of class A plexins33.A proteins that exert their biological effects through binding and crystal structure of the a1 domain of Nrp1 in complex with activating their cognate cell surface receptors. The semaphorin PlexinA2 and Sema3A shows that the Nrp1-a1 domain stabilizes family members all contain a N-terminal Sema domain, char- the complex by binding the Sema domains of both semaphorin acterized by a 7-bladed β-propeller fold, that mediates the and plexin simultaneously7. Interestingly, biochemical analyses interaction with the receptor4–10. In most semaphorins, the Sema presented in this study showed that the a1a2b1b2 domains domain is followed by a small cysteine-rich plexin-semaphorin- together are required for the semaphorin/plexin/neuropilin Integrin (PSI) domain and an immunoglobin-like (Ig-like) complex formation, but the a2b1b2 domains are not resolved in domain. Semaphorins typically form homodimers and in some the structure and therefore their roles in the complex remain cases heterodimers through interactions mediated by the Sema unclear7. In addition, the C-terminal tail region of class 3 sema- and Ig-like domains4–10. The region following the Ig-like domain phorins contributes to the complex formation by interacting with is more diverse among the semaphorin family members. A con- neuropilin25,26,35,36. Class 3 semaphorins could be cleaved at served cysteine residue in the C-terminal tail region of class multiple sites by furin or furin-like proteases, which recognize a 3 semaphorins forms an interchain disulfide to covalently link the degenerate K/R-XX-R (K, lysine; R, arginine; X, any residue) two subunits of the dimer11,12. motif37. Of particular interest, furin-cleavage of the C-terminal The best characterized receptor of semaphorins are plexins, poly-basic region of class 3 semaphorins results in a C-terminal which are classified into four classes (A, B, C, and D)1,2. Plexins arginine residue that serves as a high-affinity ligand for a pocket are type I single-pass transmembrane proteins that also contain a on the b2 domain of neuropilin37–41. Notably this pocket on the Sema domain at the N-terminus of the extracellular region, which b2 domain of neuropilin has recently been suggested to bind a C- binds the Sema domain in semaphorins. The Sema domain in terminal arginine-containing motif as a result of furin-mediated plexins is followed by multiple PSI and IPT (Integrin, Plexin, and cleavage of the spike protein of SARS-CoV-2, and thereby con- Transcription factor) domains, which connect to the transmem- tributing to viral infection42,43. brane (TM) helix. The intracellular region of all plexins contains a Taken together, neuropilin interacts with semaphorin and membrane proximal juxtamembrane region (JM), a Rho GTPase- plexin through multiple binding sites. How these multiple sites binding domain (RBD), and GTPase Activating Protein (GAP) work together in the context of the 2:2:2 semaphorin/plexin/ domain13–18. In general, the binding of dimeric semaphorin leads neuropilin assembly is not known due to lack of structure ana- to the formation of the active dimer of plexin, which activates the lyses of an intact complex. We reconstituted the complex using GAP activity of plexins specific for Rap GTPase that is essential the a1a2b1b2 domains of Nrp1, the near full-length extracellular for signaling2,13,14,19,20. The intracellular region of some plexins regions of PlexinA4 and Sema3A. We determined the structure of can interact with other proteins for signal transduction through this complex to 3.7 Å resolution by using cryo-electron micro- additional pathways or intracellular trafficking21–23. scopy (Cryo-EM) single particle reconstruction. The structure An exception to the general mechanism described above is that reveals the overall architecture of the 2:2:2 complex and the secreted class 3 semaphorins binds class A plexins very weakly, details of the multiple interfaces among the subunits that stabilize and therefore cannot induce their activation independently24–26. this large assembly. In addition, this architecture reveals pre- In this case, the transmembrane protein neuropilin (Nrp1 or viously unappreciated important roles for the long and non- Nrp2) acts as an essential co-receptor to facilitate the sema- conserved interdomain linker regions in both neuropilin and phorin/plexin interaction. Neuropilin interacts with both sema- semaphorin in the complex formation. phorin and plexin, leading to a stable 2:2:2 complex of semaphorin, plexin and neuropilin that is capable of triggering downstream signaling7. Nrp1 and Nrp2 are also type I trans- Results membrane proteins, composed of a multi-domain extracellular Structure determination of the Sema3A/PlexinA4/Nrp1 com- region, a single transmembrane helix and a short cytoplasmic tail. plex. The results from early cell-based assays have demonstrated The extracellular region contains from the N- to C-terminal that class 3 semaphorins at low- or subnanomolar concentrations region the a1, a2, b1, b2, and MAM (Meprin, A5, and Mu- can bind their plexin/neuropilin receptors and induce cell col- phosphatase) domains. The a1 and a2 domains belong to the lapse that reflect their repulsive guidance activity, indicating of Ca2+-binding CUB domain family, whereas the b1 and b2 are high-affinity interactions24–26,44. However, thorough in vitro coagulation factor V/VIII homology domains27–29. The MAM binding assays with surface plasma resonance have shown that domain exhibits a jellyroll topology and also binds Ca2+.30 The the dissociation constants between individual pairs of the Sema- extracellular domains of neuropilin play major roles in stabilizing PSI domains of Sema3A, the Sema-PSI1-IPT1-PSI2 domains of the semaphorin/plexin/neuropilin complex, while the intracel- PlexinA2 and the a1a2b1b2 domains of Nrp1 range from double- lular region interacts with proteins that contribute to downstream digit micromolar to undetectable7. The binding affinity of Plex- signaling or regulate trafficking of neuropilin2. Another inter- inA2 to Sema3A and Nrp1 both immobilized on the same sensor esting aspect is that some class 3 semaphorins require both chip is stronger (Kd value of 6 µM), but still three orders of neuropilin and the simultaneous presence of two different class A magnitude weaker than those indicated by the cell-based assays.

2 NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x ARTICLE

a Sema3A Nrp1 Sema PSI Ig-like Cys723 R770 a1 a2 b1 b2 MAM TM Signal peptide Signal peptide PlexinA4 Sema PSI1IPT1 PSI2 IPT2 PSI3 IPT3 IPT4 IPT5 IPT6 RBD Signal peptide TM JM GAP b Side view Top view

90°

c Nrp1’ PlexinA4 Sema3A’ Top view 220 Å b1’b1b1’ SemSeSema’a’ Bottom view a2’aa22’ SemSeSema a1’ b2’b2’ 110 Å a1a1 b2b2 180° SemSema’a’ aa22 SemSema b1b1 Sema3A PlexinA4’ Nrp1

90° Side view Side view b1b1 IgIg IgIg’Ig’ b1’ PSPSIPSI PSIPSPSI’ b2b2 a2a2 SemSema’a’ b2b2’b2’ a2’a2a2’

SemSeSemema 160 Å a1a1 a1’a1a1’1’

SemSemema’a’ 90° SemSema

IPTIPIPT6 IPTIPTPT6’6’ ~50 Å Extracellular Plasma membrane Intracellular

Fig. 1 Overall structure of the Sema3A/PlexinA4/Nrp1 complex. a Domain structures of mouse Sema3A, PlexinA4, and Nrp1. Colored parts indicate boundaries of constructs used in the structural analyses. b Overview of the cryo-EM map of the Sema3A/PlexinA4/Nrp1 complex. The color scheme is the same as a. c Atomic model of the Sema3A/PlexinA4/Nrp1 complex based the cryo-EM reconstruction.

This discrepancy is in part owing to the fact that the in vitro enhance the binding to plexin45. For neuropilin, we tested the binding assays used a Sema3A construct lacking the C-terminal full-length extracellular region and constructs with the MAM poly-basic tail, which is processed by furin-like proteases to domain truncated, as previous studies have shown that the MAM generate a motif with a C-terminal arginine residue that binds the domain is not directly involved binding either semaphorin or ﬁ 7,36,44,46 b1 domain of Nrp1 with high af nity (reported Kd values in the plexin . low nanomolar to micromolar range)25,38,39. In our effort of Among the many combinations of semaphorin, plexin and reconstitution of the semaphorin/plexin/neuropilin complexes, neuropilin proteins tested, the complex formed by co-expression we therefore used constructs of semaphorins containing the C- of mouse Sema3A ending at Arg770 with the A106K mutation terminal tail but truncated at one of the furin-cleavage sites, (Sema3A (21–770)-A106K), the MAM-domain truncated version equivalent to the furin-processed mature forms (Fig. 1a). We also of Nrp1 (Nrp1-a1a2b1b2) and the full-length extracellular introduced to the semaphorin constructs the mutation corre- region of PlexinA4 was stable in gel ﬁltration chromatography sponding to A106K in Sema3A, which was rationally designed to and showed relatively homogenous particles in our cryo-EM

NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x

Table 1 Data collection and model refinement statistics. membrane proximal IPT6 domain, which represents the foot tucked near the rear end of the body. The binding mode between the Sema3A dimer and the two Data collection and processing PlexinA4 molecules in the complex is similar to those in all the fi Magni cation 46,730 previous structures of semaphorin/plexin complexes, with or Voltage (kV) 300 without neuropilin (Supplementary Fig. 5)6,7,9,10,33. However, the Electron exposure (e−/Å2)50 Defocus range (µm) 1.5–2.5 cryo-EM structure provides a picture of the entire extracellular Pixel size (Å) 1.08 region of the 10-domain class A plexin in a ligand-induced active Symmetry imposed C2 dimeric complex. A superimposition based on the N-terminal Map resolution (Å) 3.7 Sema domain of the PlexinA4 extracellular region in the complex FSC threshold 0.143 with the previously reported crystal structure of PlexinA4 in the Model refinement apo-state shows that their overall ring-shapes are very similar Initial model used (PDB code) 5l5k, 4gz8 & 2qqk (Supplementary Fig. 6a)47. There are only small differences in the Model resolution (Å) 4.2 relative orientations among the domains following the Sema FSC threshold 0.5 domain, which result in a slightly more tightly curled shape of the Map sharpening B factor (Å2) −30 apo-PlexinA4 crystal structure compared with that in the cryo- Model composition EM structure. Such similar conformations of PlexinA4 obtained Non-hydrogen atoms 30,898 from completely different conditions and states highlight the Protein residues 4228 remarkable rigidity of the extracellular region of PlexinA4 despite Ligands 4 B factors (Å2) containing a total of 10 domains that are organized in a Protein 260 consecutive fashion and lack long-range interactions among non- Ligand 216 neighboring domains. The rigid ring-shape of PlexinA4 in R.m.s. deviations combination with the Sema3A/PlexinA4 binding mode leads to Bond length (Å) 0.004 close juxtaposition and roughly parallel relative orientation Bond angle (°) 0.7 between the two membrane proximal IPT6 domains from the Validation two PlexinA4 molecules in the complex (Fig. 1b, c), consistent Molprobity score 2.06 with the previous models based on the structures of the plexin Clashscore 17.4 extracellular region docked to semaphorin dimers47. The Poor rotamers (%) 1 membrane proximal IPT6 domain is connected to the transmem- Ramachandran plot brane region of plexin through a short linker (~10 residues). The Favored (%) 95.6 ~20 residues from the two copies of the linker can span the Allowed (%) 4.4 Outliers (%) 0 distance (~50 Å) between the C-termini of the two IPT domains in the dimeric complex. The architecture of the Sema3A/ PlexinA4/Nrp1 extracellular region complex is therefore poised screen (Supplementary Figs. 1, 2). We collected a cryo-EM dataset to induce the formation of the intracellular active dimer of plexin for this complex on a Titan Krios 300 kV microscope and for signaling, in a similar manner as that seen in PlexinC1 obtained a 3D reconstruction at overall resolution of 3.7 Å activated by the viral semaphorin-mimic A39R13,33. (Supplementary Fig. 2 and Table. 1). The cryo-EM map reaches high local resolutions for the Sema domains of both Sema3A and PlexinA4, whereas the C-terminal domains of both Sema3A and Role of Nrp1 in the Sema3A/PlexinA4/Nrp1 complex. Each PlexinA4 are less well resolved (Supplementary Fig. 3). All the subunit of the Sema3A dimer binds one PlexinA4 molecule to four domains (a1, a2, b1, and b2) in the Nrp1 construct are form a 2:2 complex where the two PlexinA4 molecules do not present in the reconstruction, although the density for the b1 and contact each other (Fig. 1). The 2:2 Sema3A/PlexinA4 complex b2 domains is weaker than that of a1 and a2 domains. Model can also be viewed as two Sema3A/PlexinA4 heterodimers, which building was initiated by docking available crystal structures of are held together by interactions between the two Sema3A various parts of the complex, followed by manual building and molecules as well as the interactions contributed by Nrp1. The a1 real-space refinement. The final model contains all the structured domain of Nrp1 wedges between the Sema3A molecule from one domains of the three proteins, but with various segments in some of the heterodimers and the PlexinA4 molecule from the second domains and many sidechains not included due to lack of clear heterodimer, helping glue the two heterodimers together. This density (See method for details). binding mode of the Nrp1-a1 domain with semaphorin and plexin is similar to that observed in the previous crystal structure of the complex of the Sema3A-Sema-PSI domains, the PlexinA2- Overall architecture of the Sema3A/PlexinA4/Nrp1 complex Sema-PSI1-IPT1-PSI2 domains and the Nrp1-a1 domain7. The and the activation mechanism of plexin. The cryo-EM structure higher resolution of the cryo-EM structure presented here (3.7 Å shows a 2-fold symmetric 2:2:2 complex of the extracellular resolution compared with the 7.0 Å resolution of the crystal regions of Sema3A, PlexinA4, and Nrp1, with an overall structure) reveals more details of the binding interfaces (see more dimension of ~220 × 160 × 110 Å (Fig. 1b, c and Supplementary below). The a2b1b2 domains of Nrp1, adopting the characteristic movie. 1). From a sideview, the overall shape of the complex triangular shape7,28, is placed on the side of Sema3A above both resembles a frog. The Sema domains of Sema3A and PlexinA4 the a1 domain and PlexinA4. The a2 domain of Nrp1 makes an and the a1 domain of Nrp1 together form the central compact interface with the Sema domain of Sema3A, which has not been part of the complex that can be considered the body of the frog. observed before (See more below). Presumably, the b1 domain The Ig-like domains from the two Sema3A subunits form the binds the C-terminal arginine motif of Sema3A, but this inter- head, while the two a2b1b2 modules of Nrp1 resemble two front action is not resolved in the cryo-EM map and therefore not limbs. The extracellular region of PlexinA4 adopts the highly included in the atomic model. curled ring-like shape as seen before34,47, resembling two rear The position of the a2b1b2 module of Nrp1 in the complex limbs that extend sideways but fold back and converge at the indicates that these domains are high above the surface of the

4 NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x ARTICLE

a Side view Top view

~80 Å ~80 Å C723 C C C-terminal R Ig-like Ig-like C-terminal R R b1 R b1

outside

MAM MAM MAM

center

center MAM >130 Å outside

Plasma membrane

b linker between b2 and MAM domains mNrp1 586 - - - - - EAPTAGPTTPNGNPVDECDDDQANCHSGTGDDFQLTGGTTVLATEKPT I I DST I QSEFPTY 646 mNrp2 595 TDSKPTVETLGPTVKSEETTTPYPMDEDATECGENCSF ------EDDKDLQLPS- 642

linker between MAM and transmembrane domains

mNrp1 812 P-TDLDKKNTEIKIDETGSTPGYEGE------GEGDKNISRKPGNVLKTLDP 856 mNrp2 803 P I SAFAGEDFKVD I PETHGGEGYEDE I DDEYEGDWSNSSSSTSGAGDPS - SGKEKSWLYTLDP 864

c linker between Ig-like domain and C723 C723 mSema3A 670 DTEHLEELLHKDDDGD -GS -K I KEMSSSMTPSQK -VWYRDFMQL INHPNLNTMD - -EFC 723 mSema3B 663 SAAQAERLARAEE------AAAPAPPGPK-LWYRDFLQLVEPGGGGGANSLRMC709 mSema3C 663 DSEMVAVVTDKW------SPW--TWAGSVRALPFHPKDILGAFSHSEMQLIN--QYC709 mSema3D 685 ENEQMENTQRAEYQEG - - - - -Q - - - VKDLLAESR - LRYKDY I Q I LSSPNFS - LD - -QYC 731 mSema3E 674 EEHKVEGMFHKDHEEERHHKMPCPPLSGMSQGTK - PWYKEFLQL I GYSNFQRVE - - EYC 729 mSema3F 667 GRDAVHAALFPPLAVS----VPP--P-PGTGPPT-PPYQELAQLLAQPEVGLIH--QYC715 mSema3G 675 AAVQLDSLFLRESRLE----EPSAWGSLASASPK-TWYKDILQLTGFANLPRVD--EYC726

linker between C723 and R770 R731, 734, 735, 737 R758 R761 R770

mSema3A 724 EQVWKRDRKQRRQRPGHSQGSSNKWKHMQESKKGRNRRTHEFE - - - - RAPRSV 772 mSema3B 710 R------PQPGH------HSVAADSRRKGRNRRMHVSELRAERGPRSAAHW 748 mSema3C 710 KDTRQQQQLGEEPQKMRGDY - - GK L KA L I NSRKSRNRRNQL PES 751 mSema3D 732 EQMWYKEKRRQR ------NKGSPKWKHMQEMKKKRNRRHHRDLDELQRSVAT 777 mSema3E 730 EKVWCTDKKRKK------LKMSPSKWKYANPQEKRLRSKAEHF---RLPRHTLLS 775 mSema3F 716 QGYWRHVPPRPRE - - - - - APGALRPPELQDQKKPRNRRHHPPDT 754 mSema3G 727 ERVWCRGVGERSGSFRGKGKQAKGKSWAGLELGKKMKSRVLAEHN - -RTPREVEAT 780

Fig. 2 Importance of the linker regions in the formation of the Sema3A/PlexinA4/Nrp1 complex. a Roles of the linkers in connecting various parts of the proteins in the tripartite complex on the cell surface. The crystal structure of the MAM domain of Nrp1 (PDB ID: 5l73) is docked between the a2b1b2 module and the plasma membrane. Thick lines represent the linker regions. The arrows in the right panel (top view) indicate that the MAM-linker region may use two alternative paths (through the center or outside of the PlexinA4 ring) to connect to the TM region of Nrp1. b Sequence alignment of the b2- MAM and MAM-TM linkers from mouse Nrp1 and Nrp2. c Sequence alignment of the tail regions of mouse class 3 semaphorins. Putative furin-cleavage sites are highlighted. plasma membrane where both PlexinA4 and Nrp1 are anchored distance (over 130 Å). The MAM domain adopts a compact through their respective transmembrane regions (Fig. 1). As structure, with the distance between its N- and C-termini of shown in Fig. 2, the region following the b2 domain of neuropilin, ~30 Å30. Therefore, the b2-MAM and MAM-TM linkers must which includes the b2-MAM linker, the MAM domain and the cover ~100 Å distance to allow the formation of the semaphorin/ MAM-TM linker, could reach the transmembrane region by plexin/neuropilin complex on the cell surface (Fig. 2a). Consistent passing through the center of the PlexinA4 ring, or either side of with this requirement, the two linkers in both mouse Nrp1 and the ring. In both scenarios, this region needs to span a large Nrp2 are rather long, containing ~100 residues in total (Fig. 2b).

NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x

A search in the Uniport database showed that the linkers in interact with semaphorin independent of neuropilin, do not have neuropilin from other species are in general of similar length. The a lysine residue at this position (Fig. 3c). linkers are predicted to be extended with no secondary structures, The interface between the PlexinA4-Sema and Nrp1-a1 and much less conserved compared to other regions of domains is rather small, suggesting a weak binding similar to neuropilin. Interestingly, the b2-MAM linker in Nrp2 is ~13- that between PlexinA2 and Nrp1, which has a reported value of residue shorter than that in Nrp1, whereas the MAM-TM linker dissociation constant of 66 µM7. The interaction between full- in Nrp2 is ~16-residue longer (Fig. 2b). These observations length plexin and neuropilin is likely stronger on the cell surface suggest that the total length of the two linkers together is where they may reach higher local concentrations and are important, while the distribution of residues in the two linkers confined to 2-dimentional space. In addition, it is possible that may be not, consistent with the idea that these linkers serve as they may form additional interactions with each other through spacers, rather than mediate specific protein–protein interactions. their transmembrane and intracellular regions. The differential lengths of the two linkers likely lead to different positions of the neuropilin MAM domain in the semaphorin/ Interface between Sema3A and Nrp1-a1. The Sema domain of plexin/neuropilin complex, suggesting that the MAM domain acts Sema3A in our Cryo-EM form a dimer virtually identical to all as a part of the spacer and its exact position is not important. This other dimeric semaphorins. The map for the PSI and Ig-like role for the MAM domain is consistent with the observations that domains of Sema3A is relatively weak, but allowed a partial it does not interact with either semaphorin or plexin7,36,44,46. model of these domains to be built. The model shows that the two Ig-like domains from each of the two Sema3A subunits tilt toward Interface between PlexinA4 and Sema3A. The binding mode each other and may contribute to the homodimerization by between the sema domains of PlexinA4 and Sema3A is highly making contacts through the surface-exposed face of the similar to that seen in the previously reported structures of the A–B–E–D β-sheet (Fig. 2a). The structures of Sema4D and semaphorin/plexin complexes, both those dependent and inde- Sema7A show similar contributions of the Ig-like domain to the pendent of neuropilin (Supplementary Fig. 5)6,7,9,10,33. The dimerization, but the Ig-like domain in the previous crystal A106K mutation in Sema3A was designed on the basis that the structures of class 3 and class 6 semaphorins is not resolved6,9,10. equivalent position in Sema6A is Lys110, which contributes to the The C-terminal tail following the Ig-like domain, which is not high affinity, neuropilin-independent binding with PlexinA2 resolved in the Cryo-EM structure, contains the cysteine (Cys723) through both hydrophobic and electrostatic effects45. Our struc- that forms the interchain disulfide that further stabilizes the ture confirms that the artificial Lys106 residue in Sema3A indeed Sema3A dimer (Fig. 2a). plays an analogous role in enhancing the interaction with Plex- The interface between the Nrp1-a1 domain and the Sema inA4 but does not alter the binding mode (Supplemental domain of Sema3A in the cryo-EM structure is similar to the Fig. 5)6,10. corresponding interface in the crystal structure of the Sema3A/ PlexinA2/Nrp1 complex (Fig. 4). In this case, however, the a1 domain packs more closely with the Sema domain in the Cryo- Interface between PlexinA4 and Nrp1. The only binding inter- EM structure. As mentioned above, a small shift of the a1 face between PlexinA4 and Nrp1 in the cryo-EM structure is domain away from plexin and towards semaphorin in the crystal mediated by the Sema domain of PlexinA4 and the a1 domain of structure would lead to a closer match with the cryo-EM Nrp1 (Fig. 3). The overall mode of this interface is similar to that structure. The Sema domain of Sema3A uses the middle segment seen in the previously reported crystal structure of the (residues 375–386) of the long loop between strands 3 and 4 in Sema3A–PlexinA2–Nrp1 complex7. However, a superimposition the fifth blade of the β-propeller to pack against one of the two β- of the two structures based on the plexin Sema domain shows sheets in Nrp1-a1 (Fig. 4b). The interface involves numerous that the a1 domain of Nrp1 in our cryo-EM structure is further hydrophobic packing interactions, mediated by residues, such as away from plexin by ~4 Å compared with the previous crystal Tyr375 and Phe386 from Sema3A and Tyr44, Pro45 and Pro73 structure (Fig. 3a). There are several clashes between the from Nrp1. The interface also involves sidechains of polar PlexinA2-Sema domain and the Nrp1-a1 domain in the crystal residues, including Arg372 from Sema3A, Asn72, His74, and structure, which could be resolved by a small shift of the a1 Arg137 from Nrp1. domain away from PlexinA2. It is also possible that the binding interfaces of the Nrp1-a1 domain with PlexinA2 and PlexinA4 are slightly different. Interface between Sema3A and Nrp1-a2. Previous structural The a1 domain of Nrp1 belongs to the Ca2+-binding CUB analyses have shown that the a1 domain and the triangle-shaped domain family, characterized by a 2-layered β-sandwich fold and a2b1b2 module in neuropilin are conformationally independent aCa2+ binding site formed by two inter-strand loops at one end from each other, owing to the flexible linker between them7,28. of the domain (Fig. 3a)27. Several positively charged residues from Consistent with this notion, the Cryo-EM structure here shows the edge of the PlexinA4-Sema domain, including Lys343, yet another relative orientation between the a1 and a2b1b2 Arg344, and Lys345, make electrostatic interactions with the domains of Nrp1. The linker between the a1 and a2 domains negatively charged outer surface (including Glu78, Glu126, and (residues 141–146) adopts a relatively extended conformation and Glu128) around the Ca2+-binding site in the Nrp1-a1 domain is oriented roughly orthogonal to the connecting β-strands in the (Fig. 3b). This charge-complementary binding interface is a a1 and a2 domains respectively (Fig. 4a). The position of the a2 typical feature of Ca2+-binding CUB domain-mediated domain is fixed by its interaction with the Sema domain of interactions27. The amine group in Lys343 in PlexinA4 is located Sema3A. The a2 domain is also a Ca2+-binding CUB domain. above the Ca2+-binding site and makes contacts with the Interestingly, the outer surface of the Ca2+-binding site in the a2 negatively charged residues at the binding site (Fig. 3b). Mean- domain interacts with Sema3A in a manner that resembles the while, the hydrophobic part of the Lys343 sidechain is partially interface between the Nrp1-a1 and PlexinA4-Sema domains buried in the groove formed by Tyr84, Tyr127, and Glu128 in (Fig. 4c). In this case, Lys497 in the Sema domain of Sema3A, Nrp1. Lys343 is conserved among class A plexins, suggesting that corresponding to Lys343 in PlexinA4, is partially buried in the they all use this residue to interact with the a1 domain of surface groove above the Ca2+-binding site and makes electro- neuropilin (Fig. 3c). In contrast, plexins of other classes, which static interactions with Glu195 and Asp250 in Nrp1-a2. Tyr208 in

6 NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x ARTICLE

a Plexin-Sema domain b

Sema domain R80 Nrp1-a1 domain

Y84

R344

K343 E78

K345

D85 E126 Y127 Ca2+

E128

a1 domain ~90°

Plexin-Sema domain Nrp1 Nrp1 PlexinA4 PlexinA2 (Cryo-EM structure) (from PDB ID: 4gza)

c K343

t mPlxA4 332 DLLFTVFSKGQ------KRKMKSLDESALCIFI 358 mPlxA1 333 EVLFTVFAQGQ------KNRVKPPKESALCLFT 359 4 Å shif mPlxA2 333 DVLFAIFSKGQ------KQYHHPPDDSALCAFP 359 mPlxA3 316 DVLFTIFSQGQ------KNRANPPRQTILCLFT 342 mPlxB1 291 DVLFAAFSSVAPPTVDWPLSASTGASGTSVLCAFP 325 Nrp1-a1 domain mPlxB2 292 RALYAVFSRD------GRSTGGPG-AGLCVFP 316 mPlxB3 287 DTLLGAFSAGT------SQAQAALCAFP 308 mPlxC1 304 DIWAGVFSAATGE------GQERRSPATTALCLFR 332 mPlxD1 368 EQFFAVFERPQ------GAPGARNAPAALCAFR 394

Fig. 3 Interface between PlexinA4 and Nrp1-a1. a Overview of the PlexinA4-Sema/Nrp1-a1 interface. For clarity, only one Nrp1-a1 domain and one PlexinA4 molecule from the 2:2:2 complex are shown. One complex of PlexinA2/Nrp1-a1 from PDB ID 4gza is superimposed for comparison. b Detailed view of the PlexinA4-Sema/Nrp1-a1 interface. c Sequence alignment of the loop region which in mouse PlexinA4 contains Lys343 that is important for interacting with Nrp1-a1.

Nrp1-a2 participates in the interaction by packing with the example, Sema3A could be cleaved at Arg735 or Arg76137.In hydrophobic portion of the Lys497 sidechain. Lys497 is conserved addition, due to the degeneracy of furin-cleavage motifs and the in semaphorins of class 3 but not of other classes (Fig. 4d), high abundance of lysine and arginine residues in the Sema3A suggesting that it is a specificity determinant in the interaction tail, there are several more potential cleavage sites, including with neuropilin, analogous to Lys343 in class A plexins. There are Arg731, Arg734, Arg737, and Arg758 (Fig. 2c). The tail of some additional contacts between a long loop (residues 196–208) Sema3A cleaved at Arg761 is sufficiently long and expected to be near the Ca2+-binding site in the a2 domain and the PSI domain able to interact with the Nrp1-b1 domain in the same manner as of Sema3A, although the weak density around this region pre- Arg770. However, the tail ending around Arg735 only contains vents more detailed description. 10–15 residues, which cannot cover the 80 Å distance. There are about 40 residues in the linker between the Ig-like domain (ending at around residue H680) and Cys723 (Fig. 2c). This linker Interaction between Nrp1-b1 and the C-terminal tail of could help one of the C-termini to reach the b1 domain binding Sema3A. The b1 and b2 domains in the a2b1b2 module of Nrp1 pocket on one of the Nrp1 molecules. However, in doing so, the are located away from Sema3A. The pocket in the b1 domain that second C-terminus of Sema3A would be dragged further away binds the C-terminal arginine residue of semaphorin is ~80 Å from the second b1 domain in the 2:2:2 complex, leading to an away from the Ig-like domain of Sema3A (Fig. 2a). There is a long asymmetric binding mode that is likely less stable. Interestingly, and presumably flexible linker between the interchain disulfide while the furin-cleavage site corresponding to Arg761 is largely (Cys723) to the C-terminal arginine residue (R770) (Fig. 2c). conserved in class 3 semaphorins, some of them (Sema3C, Assuming the disulfide bond is located on top of the Ig-like Sema3D, and Sema3F) lack the site equivalent to Arg770 domain at the two-fold symmetry axis of Sema3A, the two C- (Fig. 2c)37. In addition, the length of the C-terminal tail region terminal tails would diverge to opposite directions to reach for the varies among class 3 semaphorins. The different furin-cleavage b1 domains of the two Nrp1 molecules. The ~50-residues linker is patterns of Class 3 semaphorins and alternative usage of the sufficient to span the distance to allow the C-terminal arginine to multiple cleavage sites have been shown to play a role in reg- engage the binding pocket in the Nrp1-b1 domain. There are ulating their neuronal repulsive activity during embryonic several alternative furin-cleavage sites in the tail of class 3 sema- development37. The structural analyses presented here provide a phorins, which would generate shorter tails (Fig. 2c). For mechanistic explanation for such regulation.

NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x

a Binding pocket for tail Arg-motif c R207 b1 Ig-like

PSI Y208 Q498

K497 T496

D250 ~10° b2 E64 a2 S251 E195 Ca2+ Sema I253 a1 A252 E65

~20° b Nrp1 Nrp1 Sema3A Sema3A R137 (CryoEM structure) (from PDB ID: 4gza)

F386

Y38 d R350 K497 P73 N72 L353 mSema3A 491 AMELSTK - - QQQLY I G 504 mSema3B 489 SMQI SSK - -RQQLY I A 502 I107 mSema3C 488 TMK I SSK - - KQQLYVS 501 Y44 R377 mSema3D 508 NMELSLK - - QQQLYVG 521 mSema3E 493 SME I SSK - - RQQLY I G 506 Y375 mSema3F 491 TMT I SSK - -RQQLYVA 504 P45 H74 mSema3G 496 EME I SVK - - RQT LYVG 509 mSema4D 475 TLLLSSKKGRKFVYAG 490 P374 V373 mSema5A 461 SLQI LHS- -QSVLFVG 474 H46 mSema6A 489 GMQLDRA - - SGS L YVA 502 2+ Ca R372 mSema7A 465 A I SLDAD - -RRKLYVT 478

Fig. 4 Interfaces between Sema3A and Nrp1. a Overview of the Sema3A/ Nrp1 interfaces. For clarity, only one Nrp1 molecule and the Sema3A dimer from the 2:2:2 complex are shown. One complex of Sema3A/Nrp1-a1 from PDB ID 4gza is superimposed for comparison. b Detailed view of the interface between the Sema3A-Sema domain and the Nrp1-a1 domain. c Detailed view of the interface between the Sema3A-Sema domain and the Nrp1-a2 domain. d Sequence alignment of the loop region which in mouse Sema3A contains Lys497 that is important for interacting with Nrp1-a2.

Structure-based mutational analyses of the Sema3A/PlexinA4/ To test the interface between PlexinA4 and the Nrp1-a1 Nrp1 complex. To validate the Cryo-EM structure, we designed domain, we introduced a charge-reversal mutation to Lys343 in mutations based on the structure and examined their effects on PlexinA4, which makes key interactions with the Ca2+-binding both the complex formation and the cellular function of the site in Nrp1-a1 as mentioned above (Fig. 3b). As shown in complex. The complex formation was accessed by an in vitro Supplementary Fig. 7b, PlexinA4-K343E pulled down less pull-down assay with purified proteins where FLAG-tagged Sema3A and Nrp1 compared with PlexinA4-WT. Conversely, a PlexinA4 extracellular region was used to pull-down Sema3A charge-reversal mutation of Glu128 in Nrp1-a1, which interacts and Nrp1-a1a2b1b2 (Supplementary Fig. 7). The cellular with PlexinA4-Lys343, showed a similar detrimental effect as function was tested by using the well-established COS7 cell PlexinA4-K343E in the pull-down assay (Supplementary Fig. 7b). collapse assay, in which Sema3A treatment of COS7 cells Mutation of Arg80 in Nrp1-a1, located at the periphery of the expressing full-length PlexinA4 and Nrp1 leads to cell mor- PlexinA4/Nrp1-a1 interface, did not cause obvious decrease in the phological collapse that represents semaphorin-induced neu- complex formation (Supplementary Fig. 7c). For the collapse ronal growth cone collapse and repulsive guidance24.Normal assay, we used Sema3A (21–770) at 5 nM to treat COS7 cells cell surface expression of PlexinA4 and Nrp1 wild-type and stably expressing full-length PlexinA4 and Nrp1. The results mutants was confirmed by immunofluorescence microscopy showed that such treatment for 30 min led to ~93% collapse of (Supplementary Fig. 8). We did not test the interface between cells expressing PlexinA4-WT and Nrp1-WT (Fig. 5). In contrast, plexin and semaphorin, because this interface in the Cryo-EM upon the same Sema3A treatment, only ~2.5% cell expressing structure is consistent with those in previous studies, which PlexinA4-K343E and Nrp1-WT underwent collapse. Likewise, the have been extensively examined through mutational collapse percentage of cells expressing PlexinA4-WT and Nrp1- analyses6,9,10. The Sema3A proteins used in these experiments E128R was ~8%. These results are qualitatively consistent with do not contain the A106K mutation. the results from the pull-down assay, although the effects of the

8 NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x ARTICLE

Sema3A(21-770)-WT Sema3A(21-770)-WT Sema3A(21-770)-E65A Sema3A(21-770)-F386A Sema3A(21-770)-K497E Sema3A(21-734)-WT PlexinA4-WT PlexinA4-K343E PlexinA4-WT PlexinA4-WT PlexinA4-WT PlexinA4-WT Nrp1-WT Nrp1-WT Nrp1-WT Nrp1-WT Nrp1-WT Nrp1-WT

Sema3A(21-770)-WT Sema3A(21-770)-WT Sema3A(21-770)-WT Sema3A(21-770)-WT Sema3A(21-770)-WT Sema3A(21-770)-WT Sema3A(21-770)-WT PlexinA4-WT PlexinA4-WT PlexinA4-WT PlexinA4-WT PlexinA4-WT PlexinA4-WT PlexinA4-WT Nrp1-H74A Nrp1-E128R Nrp1-S251E Nrp1-A252E Nrp1-Δ591-610 Nrp1-Δ820-829 Nrp1-Δ591-610/820-829

b p=0.1553 100

p=0.0104

80 p=0.0006 p<0.0001 p<0.0001

60 p<0.0001

p<0.0001 Percentage of collapse 20 p<0.0001 p<0.0001 p<0.0001 p<0.0001 p<0.0001

Sema3A: (21-770) (21-770) (21-770) (21-770) (21-770) (21-734) (21-770) (21-770) (21-770) (21-770) (21-770) (21-770) (21-770) WT WT E65A F386A K497E WT WT WT WT WT WT WT WT PlexinA4: WTK343E WT WT WT WT WT WT WT WT WT WT WT

Nrp1: WT WT WT WT WT WT H74A E128R S251E A252E Δ591-610 Δ820-829 Δ591-610/ 820-829

Fig. 5 COS7 cell collapse assays for mutants of Sema3A, PlexinA4, and Nrp1. a Images of COS7 cell collapse. Cells stably expressing various combinations of PlexinA4 and Nrp1 constructs were treated with different Sema3A constructs at 5 nM for 30 min. Cells are visualized by immuno-staining for myc-tagged PlexinA4 (green). Nuclei are stained with DAPI (blue). One representative image from at least three biological repeats is shown for each group. b Quantification of cell collapse. The individual data points, mean, and s.e.m of the percentage of cell collapse from at least three biological repeats are shown (n = 4 for Sema3A-WT/PlexinA4-WT/Nrp1-WT, n = 5 for Sema3A-WT/PlexinA4-WT/Nrp1-S251E, n = 6 for Sema3A-WT/PlexinA4-WT/ Nrp1-A252E, n = 3 for other groups). Statistical significance p-values between wild-type and each mutant group were determined by two-tailed Student’s t- test and shown on the top of the bars. Scale bar, 50 µm. mutations are more pronounced in the collapse assay, suggesting and either Nrp1-S251E or Nrp1-A252E showed significantly that the latter is more sensitive. Due to the intricate multiple reduced collapse, ~59% and ~43%, respectively, which again interfaces made by each molecule in the 2:2:2 Sema3A/PlexinA4/ suggest that the collapse assay is more sensitive (Fig. 5). Nrp1 complex, individual mutations are expected to reduce but We also tested in the interface between the Nrp1-a1 domain not completely abolish the formation of the complex. Considering and the Sema domain of Sema3A (Fig. 4b). An alanine mutation this, the vastly different protein concentrations used in the two of Sema3A-Phe386, located at the center of this interface, led to assays, 5 nM Sema3A and cell surface expressed PlexinA4 and reduced formation of the Sema3A/PlexinA4/Nrp1 complex in the Nrp1 in the cell collapse assay versus µM concentrations of all pull-down assay (Supplementary Fig. 7b). The H74A mutation of three proteins in the pull-down assay, likely underlie their Nrp1 in the Sema3A/Nrp-a1 interface, however, did not cause different levels of sensitivity to the mutations. obvious reduction in the binding. Consistent with these results, We introduced similar mutations to test the Sema3A-Sema/ while the F386A mutation of Sema3A led to a dramatic reduction Nrp1-a2 interface (Fig. 4c). A charge-reversal mutation of in cell collapse to ~8%, the H74A mutation of Nrp1 had no Sema3A-Lys497 at the core of the interface reduced the complex significant effect on cell collapse (Fig. 5). formation in the pull-down assay (Supplementary Fig. 7b). Our analyses above suggested that the long linker between the Consistently, Seme3A-K497E only induced ~3% collapse of COS7 disulfide-forming Cys723 and the C-terminal arginine residue in cells expressing PlexinA4-WT and Nrp1-WT (Fig. 5). On the Sema3A acts as a required spacer for the optimal binding to Nrp1 Nrp1-a2 side of this interface, Ser251 and Ala252 were mutated. (Fig. 2). To test this hypothesis, we compared the complex The A252E mutant showed clear a detrimental effect in the pull- formation by two Sema3A constructs, ending at Arg734 and down assay, whereas S251E did not seem to have an effect Arg770, respectively. The results from the pull-down experiments (Supplementary Fig. 7c). COS7 cells expressing PlexinA4-WT showed that Sema3A (21–770) pulled down more PlexinA4 and

NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x

Nrp1 than Sema3A(21–734) (Supplementary Fig. 7b). Consis- conformations are not compatible with the complex formation or tently, in the COS7 cell collapse assays, while Sema3A (21–770) activation of PlexinA1 (Supplementary Fig. 6b). These observations induced ~93% collapse of cells expressing PlexinA4 and Nrp1, suggest that the conformational flexibility in the extracellular region Sema3A (21–734) at the same concentration only induced ~4% of different plexin family members may play a role in regulating collapse (Fig. 5). These results are consistent with a previous their activation. Along this line, the relatively short linker (10–15 study showing that, among the various forms of Sema3A, the one residues) between the IPT6 and transmembrane region of class A ending at Arg770 possessed the highest repulsive guidance plexin may allow the transmembrane region to sense the difference activity for cultured sympathetic ganglia from chick embryos37. in the distance between the two copies of the IPT6 domains in the As described above, the b2-MAM linker, MAM and MAM-TM 2:2:2 semaphorin/plexin/neuropilin complexes, thereby finely tuning linker regions in neuropilin do not directly mediate interactions the formation of the intracellular active dimer of plexin. This type of in the extracellular region complex, but serve as necessary spacers regulation in other receptors such as RET, the EGF receptor and c- for the formation of the complex of the full-length proteins on the Kit, has been shown to lead to differences in strength or duration of cell surface. We therefore used the COS7 cell collapse assay to signaling, and in some cases biased signaling where different examine the effects of deletions in these regions of Nrp1. We downstream pathways are activated to ultimately drive qualitatively designed a mutant with 20-residue deletion in the b2-MAM distinct biological outcomes48–51. Similar regulation of signaling in linker (Δ591–610), and a mutant with 10-residue deletion in the plexin may allow closely related plexin family members to carry out MAM-TM linker (Δ820–829). A third mutant was generated by different functions, despite their overlapping ligand-binding combining these two deletions. The results show that these link specificities. deletions reduced the percentage of cell collapse to ~50%, The Cryo-EM structure presented here clarifies the roles for the significantly lower than that of the wild-type (Fig. 5). Among five domains and the interdomain linkers of neuropilin in the all the three deletion mutants, the remaining ~70–90 residues in formation of the semaphorin/plexin/neuropilin complex. Inter- the two linkers in an extended conformation can span the 100 Å estingly, in both neuropilin and semaphorin, long flexible linkers distance required for the complex formation on the surface. are required for serving as spacers for the formation of the 2:2:2 However, the entropic cost for the mutants to do so is likely complex. Such long linkers in proteins are often neglected in increased, which may underlie their reduced but not abolished structural and functional studies because they are unstructured ability to mediate cell collapse. and nonconserved in sequence. Our analyses suggest that serving Together, the results from the pull-down and cell collapse as spacers might be a common function of long linkers, especially assays validate our cryo-EM structure and identify many residues in large multi-protein assemblies. Related to this point, the MAM that are key to the formation and signaling of the Sema3A/ domain in neuropilin is not directly involved in binding of PlexinA4/Nrp1 complex. semaphorin or plexin, but truncation of this domain abolished the ability of neuropilin in mediating semaphorin-induced neuronal growth collapse24,36,44,46. Early co-immunoprecipitation Discussion experiments suggested that the MAM domain regulate signaling While interactions of semaphorin with the plexin receptor and by mediating dimerization or oligomerization36,44,46. However, neuropilin co-receptor have been extensively investigated before, more recent structural and biophysical analyses have provided the Cryo-EM structure here provide a near complete view of the strong evidence for lack of dimerization of the MAM domain28,30. 2:2:2 extracellular region complex of these three large multi- Consistent with these results, our structure shows that the two domain proteins. The overall architecture of the 2:2:2 complex, MAM domains are placed on opposite sides of the dictated by multiple relatively weak interfaces contributed by each 2:2:2 semaphorin–plexin–neuropilin complex, unlikely to interact component, arranges the two plexin molecules in a way that can with each other. However, the MAM domain in neuropilin is a promote their activation and signaling (Fig. 6). The order in part of the linker-MAM-linker spacer that is required for the which the three proteins assemble into the 2:2:2 complex could proper formation of the 2:2:2 semaphorin/plexin/neuropilin vary as shown in Fig. 6. The heterodimerization of both sema- complex on the cell surface, which at least in part accounts for the phorin and plexin could substantially diversify the composition of functional importance of this domain in neuropilin. A remaining the 2:2:2 complexes, which may gain addition structural features question is whether the linker-MAM-linker region in neuropilin and generate different signaling outputs8,31,32. The ring-shape of connects to the transmembrane region through the center or the 10-domain extracellular region of class A plexins is essential outside of the ring of the plexin extracellular region (Fig. 2a). This for this activation mechanism, as its curvature allows the two question also pertains to how the transmembrane regions of copies of the membrane proximal IPT6 domain to converge and plexin and neuropilin are organized and whether they form thereby induce the formation of the active dimer of the intra- specific interactions in the 2:2:2 semaphorin/plexin/neuropilin cellular region13,34,47. The ring-shape has also been shown to complex. A structure of the complex of full-length semaphorin, mediate the formation of the ligand-independent dimer of class A plexin, and neuropilin is required for addressing these interesting plexins that prevents spontaneous activation in the absence of the mechanistic questions. ligand by keeping the intracellular region in the monomeric state 47 (Fig. 6) . Methods As mentioned above, the ring-shape of PlexinA4 appears quite Protein expression and purification. For recombinant expression of the rigid in both the apo-state and the 2:2:2 complex. However, three extracellular region of PlexinA4, the segment of mouse PlexinA4 cDNA crystal structures of the full-length extracellular region of PlexinA1 encoding residues 33–1226 (excluding the N-terminal signal peptide) was display substantial variations in the ring-shape33,47. Docking one of subcloned into a modified pETZ-BM vector as described previously33,which these PlexinA1 structures (PDB ID: 5l59) to the cryo-EM structure includes the N-terminal signal peptide from alkaline phosphatase and a C- terminal His8-tag (Supplementary Table 1). N-terminal FLAG-tagged PlexinA4 based on superimposition of theSemadomainofPlexinA1and were generated by inserting the FLAG-encoding sequence immediately fol- PlexinA4 shows that this conformation of PlexinA1 can form a 2:2:2 lowing the sequence peptide region through PCR reactions. The constructs for active complex similar to that of PlexinA4 (Supplementary Fig. 6b). mouse Nrp1-a1a2b1b2 (residues 22–588) and Nrp1-a1a2b1b2MAM (residues 22–855) used the same pETZ-BM vector. For expression of mouse Sema3A with However, the same modelling of the other two PlexinA1 crystal a C-terminal arginine residue, representing furin-cleaved mature isoforms, structures (PDB IDs: 5l56 and 5l5c) results in severe clashes between segments encoding residues 21–734 or 21–770 were cloned into another pETZ- the two IPT6 domains in the complex, suggesting that these BM vector that contains both the signal peptide from alkaline phosphatase and

10 NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x ARTICLE

semaphorin/ R neuropilin complex

Neuropilin

Extracellular Plasma membrane Intracellular

plexin Semaphorin-induced inhibitory dimer active 2:2:2 complex plexin/ neuropilin complex

Fig. 6 Cartoon model of the regulatory mechanisms of class A plexins. Prior to ligand binding, class A plexins form the inhibitory dimer that prevents the formation of the intracellular active dimer (left panel). It is possible that neuropilin uses its a1 domain to bind plexin, disrupting the inhibitory dimer and priming plexin for activation (lower panel). Neuropilin could bind the semaphorin dimer on its own, considering the strong interaction between the two (upper panel). Ultimately, the semaphorin dimer induces the formation of the 2:2:2 complex, which in turn induces the intracellular active dimer of plexin for signaling (right panel). Whether and how the transmembrane region of plexin and neuropilin interact in the 2:2:2 complex is unclear. a His6-tag at the N-terminus. Internal furin-cleavage sites in Sema3A, including at much higher level than Sema3A (21–770)-A106K and PlexinA4, lead to a large 551RRTRR555, 731RKQRR735,and758RNRR761 were mutated to 551ARTRA555, peak of Nrp1 in addition to the complex peak in the gel filtration run 731AAQAA735,and758ANRA761, respectively, to prevent cleavage and pro- (Supplemental Fig. 1). duction of truncated protein. All mutations were generated using PCR-based mutagenesis. Pull-down assay For expression of the individual PlexinA4, Sema3A, and Nrp1 proteins, the . The wild-type or the K343E mutant of FLAG-tagged PlexinA4 respective plasmids were transfected using polyethylenimine (PEI) into Expi293F (20 µg) was incubated with 50 µl anti-FLAG antibody conjugated resin (Sigma, cells cultured in suspension in Expi293 Expression medium (Thermo-Fisher, Cat#2220) in a buffer containing 20 mM HEPES (pH 7.5),150 mM NaCl and 1 mM CaCl at 4 °C for 30 min. Various pairs of Nrp1 (wild-type or mutants) and Cat#A1435101), with 1000 µg DNA and 4 ml PEI at 1 mg/ml for 1 L cells. For co- 2 Sema3A (wild-type or mutants), 30 µg each, were added and incubated at 4 °C for expression of the three proteins, the three plasmids at 1:1:1 weight ratio were an additional hour. In parallel, anti-FLAG resin without FLAG-tagged PlexinA4 transfected. Twelve hours post transfection, sodium butyrate with a final bound were incubated with wild-type Sema3A or Nrp1 as controls to ensure that concentration of 5 mM were added to Expi293F cells to enhance protein they do not interact with the resin nonspecifically. Resin was pelleted by cen- expression. After 7 days expression, medium was collected and exchanged to a trifugation, washed with the buffer containing 20 mM HEPES (pH 7.5), 150 mM buffer containing 25 mM Tris (pH 8.0), 500 mM NaCl, and 25 mM imidazole. The fi fi fi NaCl, and 1 mM CaCl2 three times. The samples were boiled and resolved by SDS- expressed proteins were rst puri ed by Ni-NTA af nity beads using standard PAGE. Protein was stained with Coomassie blue R250. protocols. Proteins were further purified by a Superose 6 10/300 gel filtration column (GE healthcare) with a buffer containing 20 mM HEPES (pH 7.5) and 150 mM NaCl and 1 mM CaCl2. Fractions containing the target protein or protein COS7 cell collapse assay. The cDNA of full-length mouse PlexinA4 with a C- complexes were pooled and concentrated. Purified proteins were directly used for terminal c-myc tag were subcloned into a Lentiviral vector pTY with a puromycin- preparing cryo-EM grids or stored at −80 °C for future use. Purified Sema3A resistance marker (Clontech, Mountain View, CA). The cDNA of full-length Nrp1 (21–770)-A106K, PlexinA4, and Nrp1-a1a2b1b2 could form the complex that with a C-terminal FLAG-tag were inserted into a Lentiviral vector pTY with a remained intact in gel filtration chromatography. Complexes formed with shorter blasticidin S-resistance marker. The mutants of PlexinA4 and Nrp1 were generated versions of Semaphorin were less stable and tended to dissociate in gel filtration in these constructs by PCR-based methods. The plasmids were transfected into chromatography. Sema3A (21–770)-A106K, PlexinA4, and Nrp1-a1a2b1b2 were HEK293T cells (ATCC, Cat#3216) to generate Lentiviral vectors, which were then co-expressed for convenience as well as taking advantage of the mutual used to infect COS7 cells (ATCC, catalogue #CRL-1651). COS7 cells were cultured stabilization of the proteins when the complex was formed as soon as they were in DMEM medium (Invitrogen) supplemented with 1% (V/V) penicillin/strepto- secreted into the culture medium. During co-expression, Nrp1-a1a2b1b2 expressed mycinb and 10% (V/V) FBS with 5% CO2 at 37 °C. Puromycin (1 ug/ml) and

NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications 11 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x

blasticidin S (5 ug/ml) were used together to select for COS7 cells co-expressing Reporting summary. Further information on research design is available in the Nature PlexinA4 and Nrp1. Cell surface expression of these proteins was confirmed with Research Reporting Summary linked to this article. immuno-staining and fluorescence microscopy. Myc-tagged PlexinA4 was detected by using a rabbit anti-Myc antibody (Cell Signaling technology, Cat#2278 S; 250× dilution) and an Alexa-488-conjugated anti-Rabbit IgG secondary antibody (Life Data availability Technologies, Cat#A111034; 1000× dilution). FLAG-tagged Nrp1 was detected The atomic coordinates and cryo-EM map have been deposited into the RCSB (entry ID: with a mouse anti-FLAG antibody (Bimake, Cat#A5712; 250× dilution) and an 7M0R) and EMD database (entry ID: EMD-23613), respectively. All the relevant data are anti-Mouse IgG secondary antibody conjugated with the Cy3 fluorophore (Invi- available from the authors. Source data are provided with this paper. trogen, Cat#A10521; 1000× dilution). Nuclei were stained with DAPI (4,6-diami- dion-2-phenylindole). A DeltaVision Core microscope was used to image cells. Received: 3 March 2021; Accepted: 30 April 2021; COS7 co-expressing various combinations of the wild-type or mutants of PlexinA4 and Nrp1 were seeded in six-well plates with 1 × 106 cells per well and cultured for 12 h. To induce collapse, cells were incubated with Sema3A (21–770) wild-type or mutants at a concentration of 5 nM for 30 min at 37 °C. Cells were then washed with PBS buffer and fixed in 4% paraformaldehyde. Cells were stained for PlexinA4 with an anti-myc antibody (Cell Signaling technology, Cat#2276 S; 5000× dilution) and Alexa-488-coupled anti-mouse IgG secondary antibody References (Thermo-Fisher, Cat#A11029; 1000× dilution). Nuclei were stained with DAPI. 1. Tran, T. S., Kolodkin, A. L. & Bharadwaj, R. Semaphorin regulation of cellular Images were taken at ×200 magnification with a ZOE fluorescent cell imager (Bio- morphology. Annu. Rev. Cell Dev. Biol. 23, 263–292 (2007). Rad). Collapsed cells displayed reduced area and star-like morphology, in contrast 2. Hota, P. K. & Buck, M. Plexin structures are coming: opportunities for to the characteristic large and round shape of normal uncollapsed COS7 cells. multilevel investigations of semaphorin guidance receptors, their cell signaling Counting of collapsed cells was carried out in blind. For each group, at least three mechanisms, and functions. Cell Mol. Life Sci. 69, 3765–3805 (2012). biological repeats were done. The number of cells counted in each experiment for – 3. Yoo, S. K. et al. Plexins function in epithelial repair in both Drosophila and different groups was in the range of 379 3020. Plots of percentage of collapse and fi 7 calculation of p-values were done in Graphpad Prism 9. zebra sh. Nat. Commun. , 12282 (2016). 4. Antipenko, A. et al. Structure of the semaphorin-3A receptor binding module. Neuron 39, 589–598 (2003). Cryo-EM data collection 5. Love, C. A. et al. The ligand-binding face of the semaphorins revealed by the . The Sema3A/PlexinA4/Nrp1 complex at 2 mg/ml was 10 – applied to a glow-discharged Quantifoil R1.2/1.3 300-mesh gold holey carbon grid high-resolution crystal structure of SEMA4D. Nat. Struct. Biol. , 843 848 (Quantifoil, Micro Tools GmbH, Germany). The grid was blotted under 100% (2003). humidity at 4 °C and plunged into liquid ethane with a Mark IV Vitrobot (FEI). 6. Janssen, B. J. et al. Structural basis of semaphorin-plexin signalling. Nature 467 – The serial EM software was used to control a 300 kV Titan Krios microscope (FEI) , 1118 1122 (2010). with a K3 Summit direct electron detector (Gatan) for Cryo-EM data collection52. 7. Janssen, B. J. et al. Neuropilins lock secreted semaphorins onto plexins in a The slit width of the GIF-Quantum energy filter was set to 20 eV. Images were ternary signaling complex. Nat. Struct. Mol. Biol. 19, 1293–1299 (2012). recorded in the super-resolution counting mode, dose-fractioned into 30 frames 8. Rozbesky, D. et al. Diversity of oligomerization in Drosophila semaphorins with the dose rate of 1.6 e−/Å/frame. The magnification was set to equivalent of the suggests a mechanism of functional fine-tuning. Nat. Commun. 10, 3691 pixel size of 1.08 Å. (2019). 9. Liu, H. et al. Structural basis of semaphorin-plexin recognition and viral mimicry from Sema7A and A39R complexes with PlexinC1. Cell 142, 749–761 Cryo-EM Image processing and 3D reconstruction. The image processing and (2010). 3D reconstruction processes are outlined in Supplementary Fig. 2. Movie frames 10. Nogi, T. et al. Structural basis for semaphorin signalling through the plexin were binned by the factor 2, motion corrected and dose weighted by using receptor. Nature 467, 1123–1127 (2010). Motioncorr2 1.153. GCTF 1.06 was used for CTF correction54. The rest of the 11. Klostermann, A., Lohrum, M., Adams, R. H. & Puschel, A. W. The image processing was carried out using RELION 3.155. The initial round of particle chemorepulsive activity of the axonal guidance signal semaphorin D requires picking was carried out using the Laplacian of Gaussian blob detection imple- dimerization. J. Biol. Chem. 273, 7326–7331 (1998). mented in Relion. Particles were extracted and subjected to several rounds of 2D 12. Koppel, A. M. & Raper, J. A. Collapsin-1 covalently dimerizes, and classification. A subset of particles from good 2D classes were selected for another dimerization is necessary for collapsing activity. J. Biol. Chem. 273, around of 2D classification. Nine good 2D classes averages from this step were then 15708–15713 (1998). used for reference-based particle autopicking in Relion. Particles were then 13. Wang, Y., Pascoe, H. G., Brautigam, C. A., He, H. & Zhang, X. Structural basis extracted with a binning factor of 4 and subjected to three rounds of 2D classifi- for activation and non-canonical catalysis of the Rap GTPase activating cation. Particles in good 2D classes were selected and subjected to 3D classification. protein domain of plexin. elife 2, e01279 (2013). The initial model was generated by combining the crystal structures of the partial 14. Wang, Y. et al. Plexins are GTPase-activating proteins for Rap and are Sema3A/PlexinA2/Nrp1 complex (PDB ID: 4gza) and the apo-full-length extra- activated by induced dimerization. Sci. Signal. 5, ra6 (2012). cellular region of PlexinA4 (PDB ID: 5l5k), and converting it to a density map low- 15. He, H., Yang, T., Terman, J. R. & Zhang, X. Crystal structure of the plexin A3 fi pass ltered to 30 Å resolution. A total of 137,836 particles from two good 3D intracellular region reveals an autoinhibited conformation through active site fi classes were re-extracted to the original pixel size and subjected to 3D re nement sequestration. Proc. Natl Acad. Sci. USA 106, 15610–15615 (2009). the C2 symmetry. The resulted 3D reconstruction reached 4.0 Å overall resolution, 16. Tong, Y. et al. Structure and function of the intracellular region of the plexin- but the density for the peripheral parts of the complex was much weaker than the 284 – fi fi b1 transmembrane receptor. J. Biol. Chem. , 35962 35972 (2009). center portion. A round of 3D classi cation without image alignment of the re ned 17. Tong, Y. et al. Binding of Rac1, Rnd1, and RhoD to a novel Rho GTPase particles led to one class showing stronger density for the peripheral parts. Particles interaction motif destabilizes dimerization of the plexin-B1 effector domain. J. in this class (26741 particles) were selected and subjected to further 3D refinement, Biol. Chem. 282, 37215–37224 (2007). CTF refinement, leading to a final 3D reconstruction of resolution 3.7 Å (Table 1 18. Bell, C. H., Aricescu, A. R., Jones, E. Y. & Siebold, C. A dual binding mode for and Supplemental Fig. 3). Resolution was estimated by applying a soft mask around RhoGTPases in plexin signalling. PLoS Biol. 9, e1001134 (2011). the protein density. The Fourier Shell Correlation (FSC) 0.143 criterion was used. Local resolution was calculated in Relion (Supplemental Fig. 3). 19. Pascoe, H. G., Wang, Y. & Zhang, X. Structural mechanisms of plexin signaling. Prog. Biophys. Mol. Biol. 118, 161–168 (2015). 20. Seiradake, E., Jones, E. Y. & Klein, R. Structural perspectives on axon 32 – Model building, refinement, and validation. Model building was initiated by guidance. Annu. Rev. Cell Dev. Biol. , 577 608 (2016). rigid-body docking of individual domains from the crystal structures of the mouse 21. Shang, G. et al. Structure analyses reveal a regulated oligomerization PlexinA4 extracellular region (PDB ID: 5l5k), Sema3A (PDB ID: 4gz8), and the mechanism of the PlexinD1/GIPC/myosin VI complex. elife https://doi.org/ a1a2b1b2 domains Nrp1 (PDB ID: 2qqk). Manual building was carried out using 10.7554/eLife.27322 (2017). the program Coot 0.8956. The final model does not include sidechain for many 22. Pascoe, H. G. et al. Secondary PDZ domain-binding site on class B plexins fi 112 residues, the C-terminal tail of Sem3A and several interdomain loops of the three enhances the af nity for PDZ-RhoGEF. Proc. Natl Acad. Sci. USA , – proteins due to lack of clear density. The model was refined by using the real-space 14852 14857 (2015). refinement module in the Phenix package (V1.8)57. Restraints on secondary 23. Burk, K. et al. Post-endocytic sorting of Plexin-D1 controls signal transduction structure, backbone Ramachandran angels, residue sidechain rotamers were used and development of axonal and vascular circuits. Nat. Commun. 8, 14508 during the refinement to improve the geometry of the model. MolProbity 4.5 as a (2017). part of the Phenix validation tools was used for model validation (Table. 1)58. 24. Takahashi, T. et al. Plexin-neuropilin-1 complexes form functional Figures were generated in PyMOL 2.3 (Schrödinger, LLC, 2015, the PyMOL semaphorin-3A receptors. Cell 99,59–69 (1999). Molecular Graphics System) or Chimera 1.359. Sequence alignments are rendered 25. He, Z. & Tessier-Lavigne, M. Neuropilin is a receptor for the axonal with JalView 2.11.1.360. chemorepellent Semaphorin III. Cell 90, 739–751 (1997).

12 NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23541-x ARTICLE

26. Kolodkin, A. L. et al. Neuropilin is a semaphorin III receptor. Cell 90, 753–762 53. Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam-induced motion (1997). for improved cryo-electron microscopy. Nat. Methods 14, 331–332 (2017). 27. Gaboriaud, C. et al. Structure and properties of the Ca(2+)-binding CUB 54. Zhang, K. Gctf: Real-time CTF determination and correction. J. Struct. Biol. domain, a widespread ligand-recognition unit involved in major biological 193,1–12 (2016). functions. Biochem. J. 439, 185–193 (2011). 55. Zivanov, J. et al. New tools for automated high-resolution cryo-EM structure 28. Appleton, B. A. et al. Structural studies of neuropilin/antibody complexes determination in RELION-3. elife https://doi.org/10.7554/eLife.42166 (2018). provide insights into semaphorin and VEGF binding. EMBO J. 26, 4902–4912 56. Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development (2007). of Coot. Acta Crystallogr. 66, 486–501 (2010). 29. Lee, C. C., Kreusch, A., McMullan, D., Ng, K. & Spraggon, G. Crystal structure 57. Adams, P. D. et al. PHENIX: a comprehensive Python-based system for of the human neuropilin-1 b1 domain. Structure 11,99–108 (2003). macromolecular structure solution. Acta Crystallogr. 66, 213–221 (2010). 30. Yelland, T. & Djordjevic, S. Crystal structure of the Neuropilin-1 MAM 58. Chen, V. B. et al. MolProbity: all-atom structure validation for domain: completing the Neuropilin-1 ectodomain picture. Structure 24, macromolecular crystallography. Acta Crystallogr. 66,12–21 (2010). 2008–2015 (2016). 59. Pettersen, E. F. et al. UCSF Chimera–a visualization system for exploratory 31. Kigel, B., Rabinowicz, N., Varshavsky, A., Kessler, O. & Neufeld, G. Plexin-A4 research and analysis. J. Comput. Chem. 25, 1605–1612 (2004). promotes tumor progression and tumor angiogenesis by enhancement of 60. Waterhouse, A. M., Procter, J. B., Martin, D. M., Clamp, M. & Barton, G. J. VEGF and bFGF signaling. Blood 118, 4285–4296 (2011). Jalview Version 2–a multiple sequence alignment editor and analysis 32. Sabag, A. D. et al. The role of the plexin-A2 receptor in Sema3A and Sema3B workbench. Bioinformatics 25, 1189–1191 (2009). signal transduction. J. Cell Sci. 127, 5240–5252 (2014). 33. Kuo, Y. C. et al. Cryo-EM structure of the PlexinC1/A39R complex reveals inter-domain interactions critical for ligand-induced activation. Nat. Acknowledgements Commun. 11, 1953 (2020). Cryo-EM data were collected at the University of Texas Southwestern Medical Center 34. Suzuki, K. et al. Structure of the Plexin ectodomain bound by Semaphorin- (UTSW) Cryo-Electron Microscopy Facility, funded in part by the Cancer Prevention mimicking antibodies. PLoS ONE 11, e0156719 (2016). and Research Institute of Texas (CPRIT) Core Facility Support Award RP170644. We 35. Feiner, L., Koppel, A. M., Kobayashi, H. & Raper, J. A. Secreted chick thank Dr Daniel Stoddard and Jose Martinez Diaz for facility access. We thank Hongtao semaphorins bind recombinant neuropilin with similar affinities but bind Yu and the Structural Biology Lab at UTSW for equipment usage. This work is supported different subsets of neurons in situ. Neuron 19, 539–545 (1997). in part by grants from the National Institutes Health (R35GM130289 to X.Z. and 36. Giger, R. J. et al. Neuropilin-2 is a receptor for semaphorin IV: insight into the R01GM136976 to X.-C.B.), the Welch foundation (I-1702 to X.Z. and I-1944 to X.-C.B.), structural basis of receptor function and specificity. Neuron 21, 1079–1092 and CPRIT (RP160082 to X.-C.B.). X.-C.B. and X.Z. are Virginia Murchison Linthicum (1998). Scholars in Medical Research at UTSW. 37. Adams, R. H., Lohrum, M., Klostermann, A., Betz, H. & Puschel, A. W. The chemorepulsive activity of secreted semaphorins is regulated by furin- Author contributions dependent proteolytic processing. EMBO J. 16, 6077–6086 (1997). X.Z., G.S., and X.-C.B. conceived the project. D.L., G.S., and X.H. developed the protein 38. Vander Kooi, C. W. et al. Structural basis for ligand and heparin expression and complex formation procedures. D.L. and G.S. prepared the cryo-EM binding to neuropilin B domains. Proc. Natl Acad. Sci. USA 104, 6152–6157 samples. X.-C.B. and X.Z. collected the cryo-EM data and solved the structure. D.L. did (2007). the binding and cell-based assays. X.Z. wrote the paper with inputs from other authors. 39. Guo, H. F. et al. Mechanistic basis for the potent anti-angiogenic activity of semaphorin 3 F. Biochemistry 52, 7551–7558 (2013). 40. Christensen, C. et al. Proteolytic processing converts the repelling signal Competing interests Sema3E into an inducer of invasive growth and lung metastasis. Cancer Res. The authors declare no competing interests. 65, 6167–6177 (2005). 41. Parker, M. W., Hellman, L. M., Xu, P., Fried, M. G. & Vander Kooi, C. W. Additional information Furin processing of semaphorin 3 F determines its anti-angiogenic activity by Supplementary information regulating direct binding and competition for neuropilin. Biochemistry 49, The online version contains supplementary material 4068–4075 (2010). available at https://doi.org/10.1038/s41467-021-23541-x. 42. Cantuti-Castelvetri, L. et al. Neuropilin-1 facilitates SARS-CoV-2 cell entry Correspondence and infectivity. Science https://doi.org/10.1126/science.abd2985 (2020). and requests for materials should be addressed to X.-c.B. or X.Z. 43. Daly, J. L. et al. Neuropilin-1 is a host factor for SARS-CoV-2 infection. Peer review information Nature Communications thanks Junichi Takagi and the other Science https://doi.org/10.1126/science.abd3072 (2020). anonymous reviewer(s) for their contribution to the peer review of this work. Peer 44. Chen, H., He, Z., Bagri, A. & Tessier-Lavigne, M. Semaphorin-neuropilin reviewer reports are available. interactions underlying sympathetic axon responses to class III semaphorins. Neuron 21, 1283–1290 (1998). Reprints and permission information is available at http://www.nature.com/reprints 45. Gioelli, N. et al. A rationally designed NRP1-independent superagonist SEMA3A mutant is an effective anticancer agent. Sci. Transl. Med. https://doi. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in org/10.1126/scitranslmed.aah4807 (2018). published maps and institutional affiliations. 46. Nakamura, F., Tanaka, M., Takahashi, T., Kalb, R. G. & Strittmatter, S. M. Neuropilin-1 extracellular domains mediate semaphorin D/III-induced growth cone collapse. Neuron 21, 1093–1100 (1998). 47. Kong, Y. et al. Structural basis for Plexin activation and regulation. Neuron 91, Open Access This article is licensed under a Creative Commons 548–560 (2016). Attribution 4.0 International License, which permits use, sharing, 48. Li, J. et al. Cryo-EM analyses reveal the common mechanism and adaptation, distribution and reproduction in any medium or format, as long as you give diversification in the activation of RET by different ligands. elife https://doi. appropriate credit to the original author(s) and the source, provide a link to the Creative org/10.7554/eLife.47650 (2019). Commons license, and indicate if changes were made. The images or other third party 49. Freed, D. M. et al. EGFR ligands differentially stabilize receptor dimers to material in this article are included in the article’s Creative Commons license, unless specify signaling kinetics. Cell 171, 683–695 e618 (2017). indicated otherwise in a credit line to the material. If material is not included in the 50. Ho, C. C. M. et al. Decoupling the functional pleiotropy of stem cell factor by article’s Creative Commons license and your intended use is not permitted by statutory tuning c-Kit signaling. Cell 168, 1041–1052 e1018 (2017). regulation or exceeds the permitted use, you will need to obtain permission directly from 51. Huang, Y. et al. A structural mechanism for the generation of biased agonism the copyright holder. To view a copy of this license, visit http://creativecommons.org/ in the epidermal growth factor receptor. bioRxiv https://www.biorxiv.org/ licenses/by/4.0/. content/10.1101/2020.12.08.417006v1 (2020). 52. Mastronarde, D. N. Automated electron microscope tomography using robust prediction of specimen movements. J. Struct. Biol. 152,36–51 (2005). © The Author(s) 2021

NATURE COMMUNICATIONS | (2021) 12:3172 | https://doi.org/10.1038/s41467-021-23541-x | www.nature.com/naturecommunications 13